Nosologic imaging and its value for childhood brain tumours - Essay Example

?CRITICAL REVIEW OF NOSOLOGIC IMAGING AND ITS VALUE FOR CHILDHOOD BRAIN TUMORS Introduction A latest technique has been established to develop brain nosologic images based on magnetic resonance spectroscopic imaging (MRSI) and magnetic resonance imaging (MRI). Nosologic images give a summary of the distinct lesions and tissues presence in a sole image. This is through pixel or voxel color coding in relation to the assigned histopathological class. The technique proposed utilizes advanced methods that cuts across image processing, recognition of patterns, segments and classification of brain tumors. For better understanding of how it functions, here is an illustration. For purposes of segmentation, a brain atlas that is registered in conjunction with an abnormal tissue that is subject -specific is retrieved from magnetic resonance spectroscopic imaging (MRSI) data. Subsequently, abnormal tissue detected is categorized based on pattern recognition supervised methods. In addition to that, there is computation of class probabilities for the abnormal segmented region. The new technique in comparison to former approaches is extremely flexible. Moreover, it has the capability of exploiting spatial information resulting to nosologic images that are improved. The combination of MRSI and MRI presents a new method of producing nosologic images exhibiting high resolution. Nosologic images with high resolution represent class probabilities and tumor heterogeneity which aid clinicians in making of decisions (Luts et al 2008, p.1). MRSI as a Powerful Diagnostic Tool In the current world, magnetic resonance spectroscopic imaging (MRSI) has been proved to be a diagnostic tool that is non-invasive and remarkably powerful. For instance, its ability of detecting metabolites has been extremely constructive in routine radiologic practices. This is because, it avails essential biochemical information regarding the organism molecule under investigation. In addition to that, magnetic resonance spectroscopy data has been helpful in various techniques such as tissue segmentation. The data has played a critical role in a variety of biomedical applications such as tissue volume quantification, pathologies localization, pre-surgical diagnosis improvement, therapy planning and surgical approach optimization. These applications are significant in solving diverse segmentation problems. For better understanding of various techniques of solving segmentation problems, they have been split into various categories. These are such as, classifiers, thresholding, region growing, models of Markov random field and artificial neural networks. However, Canonical Correlation Analysis (CCA) has been proposed to be a reliable and fast technique for tissue segmentation. CCA is a technique founded on statistical method. Canonical Correlation Analysis has the capability of exploiting simultaneously the spatial and spectral information. The information characterizes the data of Magnetic Resonance Spectroscopic Imaging (MRSI). CCA is successful in the application of functional data of Magnetic Resonance Imaging (MRI). The data has been useful in map sensor, cognitive and motor functions to brain specific areas. Thus, Canonical Correlation Analysis has been adopted for processing of magnetic resonance spectroscopic imaging data for purposes of detecting regions with homogeneous tissue. The regions are such as the sample characterized tumor region. The achievement of ultimate goal is reached via the combination of magnetic resonance spectroscopic spectral-spatial provided information and a subspace signal suitable for spectrum modeling of the tissue type characteristic, whose presence might be in an investigated organ and detection is needed. Canonical Correlation Analysis through the utilization of correlation coefficient quantifies the correlation between dual variable sets, and the spectra magnitude of the data measured and subspace signal. Afterwards, there is exploitation of the coefficients for purposes of constructing nosologoic images which enables the viewing of tissues detected. This necessitates the easy interpretation of images by physicians and radiologists. Together with radiologic and clinical information, diagnosis accuracy is improved. Robustness and accuracy of canonical correlation analysis outperforms significantly correlation analysis a statistical method (Laudadio et al 2005, p.1519). Magnetic Resonance Imaging (MRI) as a Tool for Assessing Brain Tumors Magnetic resonance imaging (MRI) being a contrast-enhanced tool is essential in brain tumor anatomical assessment. In comparison with tumor segmentation that is performed manually, techniques such as classification algorithms, thresholding among others have resulted to segmentation results that are more accurate. On the contrary, several diagnostic queries still emerge. These are such as tumor type and grade. These questions are complex to address utilizing the MRI conventional way. Tissue specimen histopathological characterization is and still remains the principal standard. This is in spite of the risks associated with biopsy obtaining surgery. Currently, magnetic resonance spectroscopy (MRS) is utilized frequently. MRS gives detailed metabolic information. Hence, its increased use for non-invasive and detailed specific brain tumors evaluation. MRSI (Magnetic resonance spectroscopic imaging) gives brain metabolite maps that are quantitative thus enabling the visualization of tumors heterogeneous spatial degree both outside and inside the detectable MRI lesion. Nevertheless, Magnetic resonance spectroscopic imaging analysis and inspection of obtained spectral patterns by an individual is remarkably time-consuming. Furthermore, it needs a specified spectroscopic expertise. This makes it impractical in clinical set ups. In this set up, Magnetic resonance spectroscopic imaging (MRSI) data evaluation and automated processing in conjunction with outcomes of speedy display of maps or images are necessary for exam routine interpretation in the clinic. For this reason, there was the emergency of the term nosologic image. Nosologic image refers to an image indicating a tissue type that is explicit in a distinct color. To come up with the nosologic images, various studies combined MRSI and MRI. The combination resulted to appropriate classifiers of brain tumor. However, it treated independently nosologic image voxel and pixel resulting to the exploitation of image magnitudes and spectral information. For purposes of including spatial information, application of canonical correlation analysis is mandatory to MRSI data when developing nosologic images. Limitations of Canonical Correlation Analysis (CCA) Canonical Correlation Analysis technique utilizes a sliding window. The sliding window has the capability of cancelling out impacts from delineated voxel. In addition to that, there is lack of some interactions that are long-range as a result of a window size that is fixed. Therefore, application of random fields that are conditional to magnetic resonance spectroscopic imaging (MRSI) data has been advocated. The main reason for the proposed method is its ability of increasing individual voxel classifiers performance by at most 15% (Luts et al 2008, p.3). Metabolic Imaging and MR Spectroscopy Human brain MR spectroscopy (MRS) localized proton was reported first more than two decades ago. It is a methodology utilized by worldwide medical centers for brain tumors evaluation. Heteronuclei were utilized for studying of brain tumor. Heteronuclei were such as sodium (11Na) and Phosphorus (31P). However, recently, proton nucleuses (1H) are utilized by majority of spectroscopy studies. They are highly sensitive and easily implemented in MRI scanners that are commercialized. Brain tumors MRS early development generated a wide dimension of questions such as whether MRS could or could not help in diagnosis of tumor grade and type non-invasively. The answer to this question would help in prognosis and making of management decisions. Despite the fact that MRI is the modality that is most sensitive for brain tumor detection, it has low specificity. Moreover, several kinds of tumor, lesions may exhibit same MRI manifestation. Therefore, image diagnosis has been differentiated into two. That is low-grade tumor and high-grade tumor, or non-neoplastic and neoplastic lesions. Brain tumors under high-grade category are aggressively treated than their counterparts. This makes tumor grade preoperative diagnosis to be extremely crucial. Brain biopsy operation or alternative course of treatment can only be avoided if a lesion has been diagnosed and categorized as non-neoplastic. Brain tumor mimicking on imaging that is conventional is as a result of lesions that are non-neoplatic. These lesions are infectious. Example include abscess, demyelinating lesions and ischemic lesions. Greater challenges are posed while differentiating between non-neoplastic lesions and tumors utilizing MRI conventional way. Being a technique that is remarkably sensitive for brain lesions detection, MRI capability and specificity is limited in terms of differentiating between malignant and benign lesions. For instance, a minute T2 lesions hyperintense can be extremely challenging to distinguish from other pathologies or central cortical dysplasias. Diagnostic specificity cannot be increased via the utilization of any contrast agent. The reason behind this statement is that, processes of non-neoplastic lesions are linked to blood-brain obstacle disruption. Given the fact that tumor poses increased levels of Cho and declined NAA levels, the biggest advantage of including MRS to routine clinical examinations may be such as, diagnoses excluding or including spectroscopic patterns differently marked. On the contrary, chronically demyelinating lesions and tumor differentiation basing on MRS can be cumbersome given that the entities exhibit NAA decreased levels and Cho elevated levels. In addition to that, the entities have lactate increased levels. MRI conventional technique combination with techniques that are modern can help in improving tumor classification. The modern techniques are such as physiological imaging which focuses on MRI perfusion (Horska& Barker 2010, p.4). Classification of Brain tumor utilizing Magnetic Resonance Spectroscopy (MRS) MRS as a methodology that is in-vivo and noninvasive does not need ionizing radiation. The methodology allows obtaining of metabolites profile tissue. The MRS systematic compilation following a standard acquisition procedure has enabled the signal and statistical processing techniques to be applied in the analyzing of the metabolites contribution in the tissues of the brain. The major challenge facing the radiologist for the previous two decades is the coming up with procedures that are objective. The procedures would then help in patient brain tumors diagnosis through MRS signal automatic classification. MR spectra complexity and brain tumor classification intrinsic difficulty has triggered researchers to concentrate on Machine Learning as their main objective. In addition to that, it has a methodology that is practical for identical patterns discovering from tumor tissues acquired from MR spectra. Brain tumor life cycle classification basing on magnetic resonance spectroscopy follows the methodology of machine Learning. The methodology enables finding of solution to problems resulting from Pattern Recognition. The methodology of Machine Learning comprises of two principal phases. The phases are such as Recognition and Training phase. In training phase, classification function is adapted following a group of signals acquisition procedure. Establishing of signal retrieved features and preprocessing is done at this phase. Subsequently, a model that is adaptive is fixed, elected and evaluated. All these are performed in order to receive generalization options for new cases prediction. The completed model can be utilized in conjunction with Clinical Decision Support System for new cases prediction. In that case, the steps of feature and preprocessing extraction need to be performed prior to classification function application (Garcia-Gomez 2011, p.6). Magnetic Resonance Spectroscopy Value in Imaging of childhood Tumor Childhood tumors management has been accomplished via the utilization of MRI (magnetic resonance imaging). Researchers have shown their interests in expanding their investigation to techniques of MR. The techniques are significant in providing biology tumor in vivo information. Tissue biochemistry information is provided by MRS (magnetic resonance spectroscopy). From clinical and preclinical studies, there is evident of results that are promising. Thus, a significant role is being played by magnetic resonance spectroscopy in clinical studies. In the light of the above, MRS role has not been successfully defined and there exist data scarcity from clinical multi-centre trials. Intensities and frequencies of MRS resonances are graphically represented and measured in the spectrum of MR. The frequently utilized clinical methods are 1H spectrum and 1H MRS. These methods represents macromolecules and minute mobile metabolites biochemical profiles available in a given tissue. MRI conventional investigation can be done via the utilization of MRI standard clinical scanner for purposes of carrying out 1H MRS performance. Metabolites detection technical issues have triggered brain 1H MRS studies. However, prostate and breast studies in adults have become regular. MR conventional images are used in the picking up of tissue region from which data on 1H MRS is acquired. In comparison to childhood brain tumor, studies focusing on adult brain tumors are more rampant. Pattern recognition utilizing 1H MRS has resulted to accuracy in diagnosis. Accuracy stands at 92% and above while successful diagnosis improvement stands at 15%. This is in comparison to single MRI. Enormous multi-center projects are available for purposes of developing computer software. The computer software will be beneficial in diagnosis of tumor that is 1H MRS based. The software are such as HealthAgents, eTumor and INTERPRET. Pediatric data is inclusive of HealthAgents and eTumor projects. Extremely few researchers have based their studies on brain tumor in young children. From carried out studies, it is proved that, Cho/Cr high levels can assist in differentiating between tumors of the brain and other categories of CNS lesions. This is accomplished with an accuracy of 78%. The greatest challenge in clinical is the creation of a diagnosis that is non-invasive in tumor falling under histological sub-type category. Researchers such as Wang et al managed to classify cerebellar key types of tumor in thirty one (31) children. His success was a result of measuring Cr/Cho and NAA/Cho utilizing 1H MRS. In addition to that, he joined these measurements with recognition-pattern technique making him to achieve 85% accuracy. In comparison with another study using sixty children, a great significance was noted. For example, it was established that, there was a significant distinction in many metabolites. This was significant in histological sub-types metabolites. As a result, it was proposed that, such an outcome could helpful in carrying out of diagnosis that is non-invasive. In the two studies, the main technique utilized was data-processing. 1H MRS analysis of cerebeller tumor method was recently developed. It found its application in setting majorly clinical. Measurement of additional metabolites can be achieved using 31P MRS. In future, combining of analysis software that is sophisticated with metabolite profiling that is improved will evidently boost MRS diagnostic accuracy. This will be beneficial to those young children suffering from tumor of the brain. Nevertheless, diagnostic equipment (non-invasive) predictive accuracy needs to be determined. Accuracy determination can be done in prospective multi-center trials. Moreover, their utilization in clinical environments needs demonstration(Peet et al 2008, p.726). Conclusion Canonical Correlation Analysis (CCA) being a technique that is statistical enables the quantification of the connection between random vectors that are multivariate. Despite its limitations, its potential can be applied to data sets of MRSI. The principal reason is because of its ability of exploiting spatio-spectral biomedical data. Brain tumor automatic classification utilizing MRS has been of great significance to clinical practitioners. MRS has matured enough to offer outcomes that are practical. The discipline of machine learning has been exemplarily in designing of classifiers. The classifiers are based on MRS diagnosis of brain tumor. Documentation of brain tumor evaluation and diagnosis using MRS has been done. However, MRS acceptance as clinical routine is still a controversial issue. Procedures that are automated and robust are needed in data collection, spectra analysis and outcome display. This should be done in a manner that is timely. Sites, distinct acquisition vendors’ and techniques for analysis standardization are of great significance. In addition to that, trial multi-centers should be carefully designed. While designing of the multi-centers medicine evidence-based criterion should be complied with. If the above is not achieved, then MRS will be constrained to evaluation of tumor outside medical academic centers that are valuable. Researchers should concentrate more on studying of childhood tumor rather than adult brain tumor that has been studied several times. These will enhance the acquisition of accurate diagnosis and proper treatment. References Garcia-Gomez, J 2011.” Brain tumor classification using magnetic resonance spectroscopy”. Journal of Tumors of Central Nervous Systems, vol.3, p.5-19. Horska, A., & Barker, P 2010.” Imaging of brain tumours: MR Spectroscopy and metabolic imaging.” Journal of Neuroimaging Clin N Am, vol.20, no.3, p.293-310. Laudadio, T., Pels, P., Lathauwer, LD, Hecke, P and Huffel, SV 2005.” Tissue segmentation and classification of MRSI data using canonical correlation analysis.” Magnetic Resonace in Medicine, vol.54, p.1519-1529. Luts, J., Laudadio, T., Idema, AJ., Simonetti, AW., Heerschap, A., Vandermeulen, D., Suykens, JAK., & Huffel, SV 2009.” Nosologic imaging of the brain: segmentation and classification using MRI and MRSI.” NMR Biomed, vol.22, p.374-390. Peet, AC., Arvanitis, TN., Auer, DP., Davies, NP., Hargrave, D., Howe, FA., Jaspan, T., Leach, MO., Macarthur, D., MacPherson, L., Morgan, PS., Natarajan, K., Payne, GS., Saunders, D., & Grundy, RG 2008.” The value of magnetic resonance spectroscopy in tumor imaging.” Journal of Arch Dis Child, vol.93, no.9, p. 725-727. Read More